[Model] PP support for Mamba-like models (vllm-project#10992)

Signed-off-by: mzusman <mor.zusmann@gmail.com>

Author: mzusman

docs/source/models/supported_models.rst

@@ -128,7 +128,7 @@ Text Generation
     - FalconMamba
     - :code:`tiiuae/falcon-mamba-7b`, :code:`tiiuae/falcon-mamba-7b-instruct`, etc.
     - ✅︎
-    -  
+    - ✅︎
   * - :code:`GemmaForCausalLM`
     - Gemma
     - :code:`google/gemma-2b`, :code:`google/gemma-7b`, etc.
@@ -193,7 +193,7 @@ Text Generation
     - Jamba
     - :code:`ai21labs/AI21-Jamba-1.5-Large`, :code:`ai21labs/AI21-Jamba-1.5-Mini`, :code:`ai21labs/Jamba-v0.1`, etc.
     - ✅︎
-    - 
+    - ✅︎
   * - :code:`LlamaForCausalLM`
     - Llama 3.1, Llama 3, Llama 2, LLaMA, Yi
     - :code:`meta-llama/Meta-Llama-3.1-405B-Instruct`, :code:`meta-llama/Meta-Llama-3.1-70B`, :code:`meta-llama/Meta-Llama-3-70B-Instruct`, :code:`meta-llama/Llama-2-70b-hf`, :code:`01-ai/Yi-34B`, etc.
@@ -203,7 +203,7 @@ Text Generation
     - Mamba
     - :code:`state-spaces/mamba-130m-hf`, :code:`state-spaces/mamba-790m-hf`, :code:`state-spaces/mamba-2.8b-hf`, etc.
     -
-    -
+    - ✅︎
   * - :code:`MiniCPMForCausalLM`
     - MiniCPM
     - :code:`openbmb/MiniCPM-2B-sft-bf16`, :code:`openbmb/MiniCPM-2B-dpo-bf16`, :code:`openbmb/MiniCPM-S-1B-sft`, etc.

tests/distributed/test_pipeline_parallel.py

@@ -156,13 +156,13 @@ def iter_params(self, model_name: str):
     # "internlm/internlm-chat-7b": PPTestSettings.fast(),
     "internlm/internlm2-chat-7b": PPTestSettings.fast(trust_remote_code=True),
     "inceptionai/jais-13b-chat": PPTestSettings.fast(),
-    # TODO: Implement PP
-    # "ai21labs/AI21-Jamba-1.5-Mini": PPTestSettings.fast(),
+    "ai21labs/Jamba-tiny-dev": PPTestSettings.fast(),
     "meta-llama/Meta-Llama-3-8B": PPTestSettings.detailed(),
     "openbmb/MiniCPM-2B-sft-bf16": PPTestSettings.fast(trust_remote_code=True),
     "openbmb/MiniCPM3-4B": PPTestSettings.fast(trust_remote_code=True),
     # Uses Llama
     # "mistralai/Mistral-7B-Instruct-v0.1": PPTestSettings.fast(),
+    "state-spaces/mamba-130m-hf": PPTestSettings.fast(),
     "mistralai/Mixtral-8x7B-Instruct-v0.1": PPTestSettings.fast(tp_base=4),
     "mosaicml/mpt-7b": PPTestSettings.fast(),
     "nvidia/Minitron-8B-Base": PPTestSettings.fast(),
@@ -234,6 +234,8 @@ def iter_params(self, model_name: str):
     "OpenGVLab/InternVL2-1B",
     "microsoft/Phi-3-vision-128k-instruct",
     "fixie-ai/ultravox-v0_3",
+    # [LANGUAGE GENERATION - HYBRID ARCH]
+    "ai21labs/Jamba-tiny-dev",
 ]
 
 

vllm/config.py

@@ -27,8 +27,8 @@
     ConfigFormat, get_config, get_hf_image_processor_config,
     get_hf_text_config, get_pooling_config,
     get_sentence_transformer_tokenizer_config, is_encoder_decoder, uses_mrope)
-from vllm.utils import (GiB_bytes, cuda_device_count_stateless, get_cpu_memory,
-                        print_warning_once, random_uuid,
+from vllm.utils import (GiB_bytes, LayerBlockType, cuda_device_count_stateless,
+                        get_cpu_memory, print_warning_once, random_uuid,
                         resolve_obj_by_qualname)
 
 if TYPE_CHECKING:
@@ -284,6 +284,7 @@ def __init__(
             self._verify_tokenizer_mode()
 
         self.is_attention_free = self._init_attention_free()
+        self.is_hybrid = self._init_is_hybrid()
         self.has_inner_state = self._init_has_inner_state()
 
         if current_platform.is_neuron():
@@ -340,6 +341,10 @@ def _init_attention_free(self) -> bool:
         architectures = getattr(self.hf_config, "architectures", [])
         return ModelRegistry.is_attention_free_model(architectures)
 
+    def _init_is_hybrid(self) -> bool:
+        architectures = getattr(self.hf_config, "architectures", [])
+        return ModelRegistry.is_hybrid_model(architectures)
+
     def _init_has_inner_state(self) -> bool:
         architectures = getattr(self.hf_config, "architectures", [])
         return ModelRegistry.model_has_inner_state(architectures)
@@ -669,26 +674,51 @@ def get_num_attention_heads(self,
         num_heads = getattr(self.hf_text_config, "num_attention_heads", 0)
         return num_heads // parallel_config.tensor_parallel_size
 
-    def get_num_layers(self, parallel_config: "ParallelConfig") -> int:
+    def get_layers_start_end_indices(
+            self, parallel_config: "ParallelConfig") -> Tuple[int, int]:
         from vllm.distributed.utils import get_pp_indices
         total_num_hidden_layers = getattr(self.hf_text_config,
                                           "num_hidden_layers", 0)
         pp_rank = parallel_config.rank // parallel_config.tensor_parallel_size
         pp_size = parallel_config.pipeline_parallel_size
         start, end = get_pp_indices(total_num_hidden_layers, pp_rank, pp_size)
-        return end - start
-
-    def get_num_attention_layers(self,
-                                 parallel_config: "ParallelConfig") -> int:
-        if self.is_attention_free:
-            return 0
+        return start, end
 
-        num_layers = self.get_num_layers(parallel_config)
+    def get_num_layers(self, parallel_config: "ParallelConfig") -> int:
+        start, end = self.get_layers_start_end_indices(parallel_config)
+        return end - start
 
-        # Transformers supports layers_block_type @property
-        layers = getattr(self.hf_config, "layers_block_type",
-                         ["attention"] * num_layers)
-        return len([t for t in layers if t == "attention"])
+    def get_num_layers_by_block_type(
+        self,
+        parallel_config: "ParallelConfig",
+        block_type: LayerBlockType = LayerBlockType.attention,
+    ) -> int:
+        # This function relies on 'layers_block_type' in hf_config,
+        # for w/o this attribute, we will need to have workarounds like so
+        attn_block_type = block_type == LayerBlockType.attention
+        is_transformer = not self.is_hybrid and not self.is_attention_free
+        start, end = self.get_layers_start_end_indices(parallel_config)
+
+        if is_transformer:
+            # Handle the basic case first
+            return end - start if attn_block_type else 0
+        elif self.is_attention_free:
+            # Attention free
+            # Note that this code assumes there
+            # is only one type of attention-free block type.
+            return 0 if attn_block_type else end - start
+        else:
+            # Hybrid model
+            layers_block_type_value = getattr(self.hf_config,
+                                              "layers_block_type", None)
+            if layers_block_type_value is None:
+                raise ValueError("The model is an hybrid without a"
+                                 "layers_block_type in the hf_config,"
+                                 "cannot determine the num of "
+                                 f"{block_type.value} layers")
+
+            return sum(t == block_type.value
+                       for t in layers_block_type_value[start:end])
 
     def get_multimodal_config(self) -> "MultiModalConfig":
         """

vllm/model_executor/models/interfaces.py

@@ -363,6 +363,43 @@ def is_attention_free(
     return isinstance(model, IsAttentionFree)
 
 
+@runtime_checkable
+class IsHybrid(Protocol):
+    """The interface required for all models like Jamba that have both
+    attention and mamba blocks, indicates that 
+    hf_config has 'layers_block_type'"""
+
+    is_hybrid: ClassVar[Literal[True]] = True
+    """
+        A flag that indicates this model has both mamba and attention blocks
+        , also indicates that the model's hf_config has 
+        'layers_block_type' """
+
+
+@runtime_checkable
+class _IsHybridType(Protocol):
+    is_hybrid: ClassVar[Literal[True]]
+
+
+@overload
+def is_hybrid(model: object) -> TypeIs[IsHybrid]:
+    ...
+
+
+@overload
+def is_hybrid(model: Type[object]) -> TypeIs[Type[IsHybrid]]:
+    ...
+
+
+def is_hybrid(
+    model: Union[Type[object], object]
+) -> Union[TypeIs[Type[IsHybrid]], TypeIs[IsHybrid]]:
+    if isinstance(model, type):
+        return isinstance(model, _IsHybridType)
+
+    return isinstance(model, IsHybrid)
+
+
 @runtime_checkable
 class SupportsCrossEncoding(Protocol):
     """The interface required for all models that support cross encoding."""

vllm/model_executor/models/jamba.py

@@ -9,6 +9,7 @@
 from vllm.attention.layer import Attention
 from vllm.config import _BATCH_SIZES_TO_CAPTURE, CacheConfig, VllmConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
+from vllm.distributed.parallel_state import get_pp_group
 from vllm.model_executor.layers.fused_moe import FusedMoE
 from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm.model_executor.layers.linear import (QKVParallelLinear,
@@ -25,9 +26,12 @@
                                                     MambaCacheParams)
 from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
+from vllm.utils import LayerBlockType
 
-from .interfaces import HasInnerState, SupportsLoRA
-from .utils import maybe_prefix
+from .interfaces import HasInnerState, IsHybrid, SupportsLoRA, SupportsPP
+from .utils import (is_pp_missing_parameter,
+                    make_empty_intermediate_tensors_factory, make_layers,
+                    maybe_prefix)
 
 KVCache = Tuple[torch.Tensor, torch.Tensor]
 
@@ -281,16 +285,24 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
             org_num_embeddings=config.vocab_size,
         )
 
-        decoder_layers = []
-        for i in range(config.num_hidden_layers):
-            layer_class = ALL_DECODER_LAYER_TYPES[config.layers_block_type[i]]
-            decoder_layers.append(
-                layer_class(config,
-                            layer_idx=i,
-                            cache_config=cache_config,
-                            quant_config=quant_config,
-                            prefix=f"{prefix}.layers.{i}"))
-        self.layers = nn.ModuleList(decoder_layers)
+        def get_layer(prefix: str):
+            layer_idx = int(prefix.rsplit(".", 1)[1])
+            layer_class = ALL_DECODER_LAYER_TYPES[
+                config.layers_block_type[layer_idx]]
+            return layer_class(
+                config,
+                layer_idx,
+                cache_config,
+                quant_config=quant_config,
+                prefix=prefix,
+            )
+
+        self.start_layer, self.end_layer, self.layers = make_layers(
+            config.num_hidden_layers, get_layer, prefix=f"{prefix}.layers")
+        self.make_empty_intermediate_tensors = (
+            make_empty_intermediate_tensors_factory(
+                ["hidden_states", "residual"], config.hidden_size))
+
         self.final_layernorm = RMSNorm(config.hidden_size,
                                        eps=config.rms_norm_eps)
 
@@ -304,26 +316,34 @@ def forward(
         kv_caches: List[torch.Tensor],
         attn_metadata: AttentionMetadata,
         mamba_cache_params: MambaCacheParams,
+        intermediate_tensors: Optional[IntermediateTensors] = None,
         inputs_embeds: Optional[torch.Tensor] = None,
     ) -> torch.Tensor:
-        if inputs_embeds is not None:
-            hidden_states = inputs_embeds
+        if get_pp_group().is_first_rank:
+            if inputs_embeds is not None:
+                hidden_states = inputs_embeds
+            else:
+                hidden_states = self.get_input_embeddings(input_ids)
+            residual = None
         else:
-            hidden_states = self.get_input_embeddings(input_ids)
-        residual = None
-        for i in range(len(self.layers)):
+            assert intermediate_tensors is not None
+            hidden_states = intermediate_tensors["hidden_states"]
+            residual = intermediate_tensors["residual"]
+
+        kv_cache_index = 0
+        mamba_cache_index = 0
+        for i in range(self.start_layer, self.end_layer):
             layer = self.layers[i]
             kv_cache = None
             layer_mamba_cache_params = None
             if isinstance(layer, JambaAttentionDecoderLayer):
-                kv_cache = kv_caches[(i - self.config.attn_layer_offset) //
-                                     self.config.attn_layer_period]
+                kv_cache = kv_caches[kv_cache_index]
+                kv_cache_index += 1
             if isinstance(layer, JambaMambaDecoderLayer):
-                current_state_layer = i - (1 +
-                                           (i - self.config.attn_layer_offset)
-                                           // self.config.attn_layer_period)
+                current_state_layer = mamba_cache_index
                 layer_mamba_cache_params = mamba_cache_params.at_layer_idx(
                     current_state_layer)
+                mamba_cache_index += 1
 
             hidden_states, residual = layer(
                 positions=positions,
@@ -332,11 +352,17 @@ def forward(
                 attn_metadata=attn_metadata,
                 residual=residual,
                 mamba_cache_params=layer_mamba_cache_params)
+        if not get_pp_group().is_last_rank:
+            return IntermediateTensors({
+                "hidden_states": hidden_states,
+                "residual": residual
+            })
         hidden_states, _ = self.final_layernorm(hidden_states, residual)
         return hidden_states
 
 
-class JambaForCausalLM(nn.Module, HasInnerState, SupportsLoRA):
+class JambaForCausalLM(nn.Module, HasInnerState, SupportsLoRA, SupportsPP,
+                       IsHybrid):
     packed_modules_mapping = {
         "qkv_proj": [
             "q_proj",
@@ -368,6 +394,8 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
 
         super().__init__()
         self.config = config
+        self.vllm_config = vllm_config
+        self.model_config = vllm_config.model_config
         self.scheduler_config = scheduler_config
         self.model = JambaModel(vllm_config=vllm_config,
                                 prefix=maybe_prefix(prefix, "model"))
@@ -390,6 +418,9 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
                                                 config.vocab_size)
         self.sampler = get_sampler()
 
+        self.make_empty_intermediate_tensors = (
+            self.model.make_empty_intermediate_tensors)
+
     def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
         return self.model.get_input_embeddings(input_ids)
 
@@ -406,10 +437,8 @@ def forward(self,
                 self.scheduler_config.max_num_seqs) if self.scheduler_config
                               else max(_BATCH_SIZES_TO_CAPTURE) + 2)
 
-            layers_type = self.config.layers_block_type
-            num_mamba_layers = sum(
-                [layer_type == "mamba" for layer_type in layers_type])
-
+            num_mamba_layers = self.model_config.get_num_layers_by_block_type(
+                self.vllm_config.parallel_config, LayerBlockType.mamba)
             self.mamba_cache = MambaCacheManager(
                 self.lm_head.weight.dtype, num_mamba_layers, max_batch_size,
                 *self._get_mamba_cache_shape())
@@ -423,7 +452,7 @@ def forward(self,
                                               state_indices_tensor)
         hidden_states = self.model(input_ids, positions, kv_caches,
                                    attn_metadata, mamba_cache_params,
-                                   inputs_embeds)
+                                   intermediate_tensors, inputs_embeds)
         return hidden_states
 
     def copy_inputs_before_cuda_graphs(self, input_buffers, **kwargs):
@@ -504,8 +533,12 @@ def load_weights(self, weights: Iterable[Tuple[str,
                     continue
                 name = name.replace(weight_name, param_name)
                 # Skip loading extra bias for GPTQ models.
+
                 if name.endswith(".bias") and name not in params_dict:
                     continue
+                # Skip layers on other devices.
+                if is_pp_missing_parameter(name, self):
+                    continue
                 param = params_dict[name]
                 weight_loader = param.weight_loader
                 weight_loader(param, loaded_weight, shard_id)
@@ -520,6 +553,8 @@ def load_weights(self, weights: Iterable[Tuple[str,
                     if weight_name not in name:
                         continue
 
+                    if is_pp_missing_parameter(name, self):
+                        continue
                     name = name.replace(weight_name, param_name)
                     param = params_dict[name]
                     weight_loader = param.weight_loader
@@ -533,6 +568,8 @@ def load_weights(self, weights: Iterable[Tuple[str,
                     # Skip loading extra bias for GPTQ models.
                     if name.endswith(".bias") and name not in params_dict:
                         continue
+                    if is_pp_missing_parameter(name, self):
+                        continue
 
                     param = params_dict[name]
                     weight_loader = getattr(param, "weight_loader",